Fundamental strategies first adopted in major centers 20 years ago for the physical rehabilitation of persons with sub-acute or chronic spinal cord injury (SCI) have changed little since that time. One novel method of gait rehabilitation involves the use of an overhead support point and a harness. This body weight support' (BWS) strategy has been combined with treadmill-based gait training in several centers throughout the world, with what are claimed to be dramatic results; we also have seen marked improvement in function in our subjects trained in this manner from preliminary studies. It is speculated that this form of training may enhance output of a 'central pattern generator' of stepping movement from circuitry intrinsic to the subject's spinal cord. However, published studies of this type have been limited in their ability to form strong conclusions due to small sample size, inadequate control interventions, and/or limited outcome measures. In particular, only limited attention has been paid to the role that training-induced physical conditioning might play in mediating the functional improvements reported. This study will address these weaknesses to determine whether BWS-gait training is more effective than conventional rehabilitation therapy in improving functional gait in person's with neurologically-incomplete spinal cord injury. Two populations will be studied: persons with chronic SCI (greater than 1 year post-injury), and persons with sub-acute SCI (2 - 8 months post-injury). For the chronic SCI study, subjects will be randomly assigned to one of 3 groups: body weight support and treadmill-based training, body weight support and overground training, and conventional rehabilitation therapy. This design allows us to directly compare whether treadmill-based training, and its inherent advantage of providing highly-rhythmic input to the subject's legs, is superior to overground-based training, and its inherent advantage of allowing use of assistive devices, thereby replicating a more 'natural' training condition. Training sessions will typically last up to 1 hour/session, at a frequency of 3 sessions per week for a 13 week period. Persons with sub-acute injury will be randomized to receive either BWS-treadmill training, or conventional rehabilitation. All subjects will be evaluated with a battery of functional, metabolic and neurophysiologic measures prior to the onset of training, and during the week after training has been completed. The primary outcome measure will be average maximum overground walking velocity without body weight support but with the use of passive assistive devices. Secondary measures will concentrate on function (balance, mobility), fitness (work capacity, strength, gait efficiency), and spinal cord neurophysiology (motor conduction, reflex excitability). These studies will allow us to determine whether the functional improvements associated with BWS-based training are due to neurologic adaptation within the spinal cord, or reflect an increased work capacity secondary to fitness training; both possibilities have important implications with respect to optimizing therapy for persons with spinal cord injury.